4.0 is getting things wrong that 3.5 got right

I stripped the header down to “stick to the context” basically, and dropped temp and top_p all the way down to zero. Its like it is ignoring all of it. Has anyone else had this issue?

The temp setting appears to be aligned differently with this model … I’ve had success with .9 on GPT-4 and 1.23 on ChatGPT-turbo…


I have them set to zero right now. I am just trying to find a place where it speaks the truth about what is in the context.

I got it to speak the truth once, but I had to give it 6000 tokens worth of context.